kernel/irq/spurious.c

   1 /*
   2  * linux/kernel/irq/spurious.c
   3  *
   4  * Copyright (C) 1992, 1998-2004 Linus Torvalds, Ingo Molnar
   5  *
   6  * This file contains spurious interrupt handling.
   7  */
   8
   9 #include <linux/jiffies.h>
  10 #include <linux/irq.h>
  11 #include <linux/module.h>
  12 #include <linux/kallsyms.h>
  13 #include <linux/interrupt.h>
  14 #include <linux/moduleparam.h>
  15 #include <linux/timer.h>
  16
  17 #include "internals.h"
  18
  19 static int irqfixup __read_mostly;
  20
  21 #define POLL_SPURIOUS_IRQ_INTERVAL (HZ/10)
  22 static void poll_spurious_irqs(unsigned long dummy);
  23 static DEFINE_TIMER(poll_spurious_irq_timer, poll_spurious_irqs, 0, 0);
  24 static int irq_poll_cpu;
  25 static atomic_t irq_poll_active;
  26
  27 /*
  28  * We wait here for a poller to finish.
  29  *
  30  * If the poll runs on this CPU, then we yell loudly and return
  31  * false. That will leave the interrupt line disabled in the worst
  32  * case, but it should never happen.
  33  *
  34  * We wait until the poller is done and then recheck disabled and
  35  * action (about to be disabled). Only if it's still active, we return
  36  * true and let the handler run.
  37  */
  38 bool irq_wait_for_poll(struct irq_desc *desc)
  39 {
  40         if (WARN_ONCE(irq_poll_cpu == smp_processor_id(),
  41                       "irq poll in progress on cpu %d for irq %d\n",
  42                       smp_processor_id(), desc->irq_data.irq))
  43                 return false;
  44
  45 #ifdef CONFIG_SMP
  46         do {
  47                 raw_spin_unlock(&desc->lock);
  48                 while (irqd_irq_inprogress(&desc->irq_data))
  49                         cpu_relax();
  50                 raw_spin_lock(&desc->lock);
  51         } while (irqd_irq_inprogress(&desc->irq_data));
  52         /* Might have been disabled in meantime */
  53         return !irqd_irq_disabled(&desc->irq_data) && desc->action;
  54 #else
  55         return false;
  56 #endif
  57 }
  58
  59
  60 /*
  61  * Recovery handler for misrouted interrupts.
  62  */
  63 static int try_one_irq(int irq, struct irq_desc *desc, bool force)
  64 {
  65         irqreturn_t ret = IRQ_NONE;
  66         struct irqaction *action;
  67
  68         raw_spin_lock(&desc->lock);
  69
  70         /* PER_CPU and nested thread interrupts are never polled */
  71         if (irq_settings_is_per_cpu(desc) || irq_settings_is_nested_thread(desc))
  72                 goto out;
  73
  74         /*
  75          * Do not poll disabled interrupts unless the spurious
  76          * disabled poller asks explicitely.
  77          */
  78         if (irqd_irq_disabled(&desc->irq_data) && !force)
  79                 goto out;
  80
  81         /*
  82          * All handlers must agree on IRQF_SHARED, so we test just the
  83          * first.
  84          */
  85         action = desc->action;
  86         if (!action || !(action->flags & IRQF_SHARED) ||
  87             (action->flags & __IRQF_TIMER))
  88                 goto out;
  89
  90         /* Already running on another processor */
  91         if (irqd_irq_inprogress(&desc->irq_data)) {
  92                 /*
  93                  * Already running: If it is shared get the other
  94                  * CPU to go looking for our mystery interrupt too
  95                  */
  96                 desc->istate |= IRQS_PENDING;
  97                 goto out;
  98         }
  99
 100         /* Mark it poll in progress */
 101         desc->istate |= IRQS_POLL_INPROGRESS;
 102         do {
 103                 if (handle_irq_event(desc) == IRQ_HANDLED)
 104                         ret = IRQ_HANDLED;
 105                 /* Make sure that there is still a valid action */
 106                 action = desc->action;
 107         } while ((desc->istate & IRQS_PENDING) && action);
 108         desc->istate &= ~IRQS_POLL_INPROGRESS;
 109 out:
 110         raw_spin_unlock(&desc->lock);
 111         return ret == IRQ_HANDLED;
 112 }
 113
 114 static int misrouted_irq(int irq)
 115 {
 116         struct irq_desc *desc;
 117         int i, ok = 0;
 118
 119         if (atomic_inc_return(&irq_poll_active) != 1)
 120                 goto out;
 121
 122         irq_poll_cpu = smp_processor_id();
 123
 124         for_each_irq_desc(i, desc) {
 125                 if (!i)
 126                          continue;
 127
 128                 if (i == irq)   /* Already tried */
 129                         continue;
 130
 131                 if (try_one_irq(i, desc, false))
 132                         ok = 1;
 133         }
 134 out:
 135         atomic_dec(&irq_poll_active);
 136         /* So the caller can adjust the irq error counts */
 137         return ok;
 138 }
 139
 140 static void poll_spurious_irqs(unsigned long dummy)
 141 {
 142         struct irq_desc *desc;
 143         int i;
 144
 145         if (atomic_inc_return(&irq_poll_active) != 1)
 146                 goto out;
 147         irq_poll_cpu = smp_processor_id();
 148
 149         for_each_irq_desc(i, desc) {
 150                 unsigned int state;
 151
 152                 if (!i)
 153                          continue;
 154
 155                 /* Racy but it doesn't matter */
 156                 state = desc->istate;
 157                 barrier();
 158                 if (!(state & IRQS_SPURIOUS_DISABLED))
 159                         continue;
 160
 161                 local_irq_disable();
 162                 try_one_irq(i, desc, true);
 163                 local_irq_enable();
 164         }
 165 out:
 166         atomic_dec(&irq_poll_active);
 167         mod_timer(&poll_spurious_irq_timer,
 168                   jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
 169 }
 170
 171 static inline int bad_action_ret(irqreturn_t action_ret)
 172 {
 173         if (likely(action_ret <= (IRQ_HANDLED | IRQ_WAKE_THREAD)))
 174                 return 0;
 175         return 1;
 176 }
 177
 178 /*
 179  * If 99,900 of the previous 100,000 interrupts have not been handled
 180  * then assume that the IRQ is stuck in some manner. Drop a diagnostic
 181  * and try to turn the IRQ off.
 182  *
 183  * (The other 100-of-100,000 interrupts may have been a correctly
 184  *  functioning device sharing an IRQ with the failing one)
 185  */
 186 static void
 187 __report_bad_irq(unsigned int irq, struct irq_desc *desc,
 188                  irqreturn_t action_ret)
 189 {
 190         struct irqaction *action;
 191         unsigned long flags;
 192
 193         if (bad_action_ret(action_ret)) {
 194                 printk(KERN_ERR "irq event %d: bogus return value %x\n",
 195                                 irq, action_ret);
 196         } else {
 197                 printk(KERN_ERR "irq %d: nobody cared (try booting with "
 198                                 "the \"irqpoll\" option)\n", irq);
 199         }
 200         dump_stack();
 201         printk(KERN_ERR "handlers:\n");
 202
 203         /*
 204          * We need to take desc->lock here. note_interrupt() is called
 205          * w/o desc->lock held, but IRQ_PROGRESS set. We might race
 206          * with something else removing an action. It's ok to take
 207          * desc->lock here. See synchronize_irq().
 208          */
 209         raw_spin_lock_irqsave(&desc->lock, flags);
 210         action = desc->action;
 211         while (action) {
 212                 printk(KERN_ERR "[<%p>] %pf", action->handler, action->handler);
 213                 if (action->thread_fn)
 214                         printk(KERN_CONT " threaded [<%p>] %pf",
 215                                         action->thread_fn, action->thread_fn);
 216                 printk(KERN_CONT "\n");
 217                 action = action->next;
 218         }
 219         raw_spin_unlock_irqrestore(&desc->lock, flags);
 220 }
 221
 222 static void
 223 report_bad_irq(unsigned int irq, struct irq_desc *desc, irqreturn_t action_ret)
 224 {
 225         static int count = 100;
 226
 227         if (count > 0) {
 228                 count--;
 229                 __report_bad_irq(irq, desc, action_ret);
 230         }
 231 }
 232
 233 static inline int
 234 try_misrouted_irq(unsigned int irq, struct irq_desc *desc,
 235                   irqreturn_t action_ret)
 236 {
 237         struct irqaction *action;
 238
 239         if (!irqfixup)
 240                 return 0;
 241
 242         /* We didn't actually handle the IRQ - see if it was misrouted? */
 243         if (action_ret == IRQ_NONE)
 244                 return 1;
 245
 246         /*
 247          * But for 'irqfixup == 2' we also do it for handled interrupts if
 248          * they are marked as IRQF_IRQPOLL (or for irq zero, which is the
 249          * traditional PC timer interrupt.. Legacy)
 250          */
 251         if (irqfixup < 2)
 252                 return 0;
 253
 254         if (!irq)
 255                 return 1;
 256
 257         /*
 258          * Since we don't get the descriptor lock, "action" can
 259          * change under us.  We don't really care, but we don't
 260          * want to follow a NULL pointer. So tell the compiler to
 261          * just load it once by using a barrier.
 262          */
 263         action = desc->action;
 264         barrier();
 265         return action && (action->flags & IRQF_IRQPOLL);
 266 }
 267
 268 void note_interrupt(unsigned int irq, struct irq_desc *desc,
 269                     irqreturn_t action_ret)
 270 {
 271         if (desc->istate & IRQS_POLL_INPROGRESS)
 272                 return;
 273
 274         /* we get here again via the threaded handler */
 275         if (action_ret == IRQ_WAKE_THREAD)
 276                 return;
 277
 278         if (bad_action_ret(action_ret)) {
 279                 report_bad_irq(irq, desc, action_ret);
 280                 return;
 281         }
 282
 283         if (unlikely(action_ret == IRQ_NONE)) {
 284                 /*
 285                  * If we are seeing only the odd spurious IRQ caused by
 286                  * bus asynchronicity then don't eventually trigger an error,
 287                  * otherwise the counter becomes a doomsday timer for otherwise
 288                  * working systems
 289                  */
 290                 if (time_after(jiffies, desc->last_unhandled + HZ/10))
 291                         desc->irqs_unhandled = 1;
 292                 else
 293                         desc->irqs_unhandled++;
 294                 desc->last_unhandled = jiffies;
 295         }
 296
 297         if (unlikely(try_misrouted_irq(irq, desc, action_ret))) {
 298                 int ok = misrouted_irq(irq);
 299                 if (action_ret == IRQ_NONE)
 300                         desc->irqs_unhandled -= ok;
 301         }
 302
 303         desc->irq_count++;
 304         if (likely(desc->irq_count < 100000))
 305                 return;
 306
 307         desc->irq_count = 0;
 308         if (unlikely(desc->irqs_unhandled > 99900)) {
 309                 /*
 310                  * The interrupt is stuck
 311                  */
 312                 __report_bad_irq(irq, desc, action_ret);
 313                 /*
 314                  * Now kill the IRQ
 315                  */
 316                 printk(KERN_EMERG "Disabling IRQ #%d\n", irq);
 317                 desc->istate |= IRQS_SPURIOUS_DISABLED;
 318                 desc->depth++;
 319                 irq_disable(desc);
 320
 321                 mod_timer(&poll_spurious_irq_timer,
 322                           jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
 323         }
 324         desc->irqs_unhandled = 0;
 325 }
 326
 327 int noirqdebug __read_mostly;
 328
 329 int noirqdebug_setup(char *str)
 330 {
 331         noirqdebug = 1;
 332         printk(KERN_INFO "IRQ lockup detection disabled\n");
 333
 334         return 1;
 335 }
 336
 337 __setup("noirqdebug", noirqdebug_setup);
 338 module_param(noirqdebug, bool, 0644);
 339 MODULE_PARM_DESC(noirqdebug, "Disable irq lockup detection when true");
 340
 341 static int __init irqfixup_setup(char *str)
 342 {
 343         irqfixup = 1;
 344         printk(KERN_WARNING "Misrouted IRQ fixup support enabled.\n");
 345         printk(KERN_WARNING "This may impact system performance.\n");
 346
 347         return 1;
 348 }
 349
 350 __setup("irqfixup", irqfixup_setup);
 351 module_param(irqfixup, int, 0644);
 352
 353 static int __init irqpoll_setup(char *str)
 354 {
 355         irqfixup = 2;
 356         printk(KERN_WARNING "Misrouted IRQ fixup and polling support "
 357                                 "enabled\n");
 358         printk(KERN_WARNING "This may significantly impact system "
 359                                 "performance\n");
 360         return 1;
 361 }
 362
 363 __setup("irqpoll", irqpoll_setup);